Magistral Medium 1.2: benchmark results
Mistral's Magistral Medium 1.2 (September 18, 2025), a proprietary multimodal reasoning model with a vision encoder for text and image input, using extended chain-of-thought with a 128K context. Provider: Mistral. Released 2025-09-17. Access: API.
Aggregate Rating 1614 ± 19, rank #391 of 1640 rated models, from 106 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA LiveCodeBench | 75.03 | Pass@1 (%) | 86.5 |
| AIME 2024 | 91.8 | Score (percentage) | 80 |
| SnorkelSpatial | 44.24 | Accuracy@1 (%, 330 spatial reasoning questions) | 78.1 |
| AA AIME 2025 | 82 | Accuracy (%) | 77.7 |
| AA MMLU-Pro | 81.51 | Accuracy (%) | 75.3 |
| AA-Omniscience Hallucination Rate | 59.98 | Hallucination Rate (%) | 72.6 |
| AA Omniscience - Software Engineering (SWE) - Rust | 54 | Accuracy (%) | 68.9 |
| AA Omniscience - Software Engineering (SWE) - Julia | 16 | Accuracy (%) | 68.8 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 22 | Accuracy (%) | 68.6 |
| AA Omniscience - Software Engineering (SWE) - Go | 22 | Accuracy (%) | 68.4 |
| Wolfram LLM Benchmarking Project | 51.7 | Correct Functionality (%) | 67.3 |
| AA Omniscience - Software Engineering (SWE) - Java | 17 | Accuracy (%) | 62.8 |
Interactive version: theaggregate.ai/model?slug=magistral-medium-1-2 · How It Works · Data refreshed daily, snapshot 2026-10-13.